Structure based drug design is critically dependent on accuracy of scoring functions, and there is of significant interest to advance protein-ligand scoring functions with machine learning approaches. In this work, by judiciously expanding the training set, establishing a time-split validation set, and exploring new features related to ligand conformation stability as well as explicit mediating water molecules, we have improved robustness and applicability of machine-learning scoring functions via extreme gradient boosting (XGBoost) and Δ-Vina parameterization. The new scoring function Δ<sub>vina</sub>XGB can not only perform consistently among the top compared to classical scoring functions for the CASF-2013 benchmark with crystal structures, but also achieve significantly better scoring accuracy in a variety of test tasks that mimic real docking applications

Incorporating ligand conformation stability and explicit water molecules in machine-learning scoring functions

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.